Cressy from the Qubit | qbitai
Neutrality and outspokenness are the selling points that "Musk's version of ChatGPT" Grok has been promoting.
But recently, netizens found that its answers were becoming more and more tendentious, and they couldn't help but post complaints, and @ Musk himself.
In the spit person's own words, Grok is starting to become more and more "chatgpt-like".
Peterson, a professor of psychology at the University of Toronto and best-selling author, posted on the article that Grok has experienced a "wake-up" phenomenon like ChatGPT, and his answers have become biased.
The so-called "awakening phenomenon" refers to the fact that the responses of the large model are no longer based entirely on facts, but begin to take on people's subjective positions, or refuse to answer sensitive questions.
At the same time, some people said that they have been tracking the bias of various large models, and it turned out that Grok is not as neutral as ChatGPT.
In this regard, some netizens expressed concern - because people can only train these tools (AI) based on human descriptions and not the facts themselves.
Some netizens seconded that AI is "not intelligence", but programs that imitate human thinking, and that they are not biased in themselves, and that their developers are biased.
The series of complaints caught the attention of the Washington Post, which directly published an article saying that Grok's work "did not go as planned."
At the same time, such a performance of Grok also alarmed Musk himself.
He complained that the Internet information used to train Grok was too complicated and full of "spam information with wake-up colors", which was equivalent to indirectly admitting the claims of netizens.
At the same time, Musk also said: The current grok is just a beta version, and it will get better in the future.
The "anti-wake" initiative instead "woke up" grok
Musk said that the purpose of developing Grok was to create a humorous and "anti-wake" chatbot.
At the same time, Musk also criticized ChatGPT for being "over-woken up".
Before the emergence of Grok's development team, XAI, Musk even accused OpenAI of "training AI to lie" and announced that he would build ChatGPT's challenger "TruthGPT".
Eventually, "TruthGPT" was named Grok, and it was advertised as "the ability to answer tough questions that other AIs refuse to answer."
Unlike other models, Grok is trained on real-time data, so it has the most up-to-date information on real-world problems.
And the XAI team doesn't filter this information, trying to keep a neutral stance while grok is "poisonous".
Although the original intention of this approach is to be neutral, many netizens believe that the reason why grok's answers have begun to appear tendentious recently may be related to these training data-
The Internet, which carries AI training data, is itself "woken up".
This also coincides with Musk's thinking that although Grok does not screen the training content, the content itself is biased enough to deviate from a neutral position.
However, Musk also said that Grok will get better in the future, and netizens also expressed support and encouragement for him and Grok.
Are you looking forward to grok's future performance?
Reference link: [1].